An AI system to help scientists write expert-level empirical software
A new AI system called Empirical Research Assistance (ERA) has been developed to aid scientists in creating expert-level empirical software. By utilizing a Large Language Model and Tree Search techniques, ERA can systematically enhance software quality and explore complex research ideas. The system has shown promising results in various fields, including bioinformatics and epidemiology, significantly outperforming human-developed methods.
- ▪ERA is designed to address the slow, manual creation of software for computational experiments.
- ▪The AI system has discovered 40 novel methods for single-cell data analysis that surpassed top human-developed methods.
- ▪In epidemiology, ERA generated 14 models that outperformed the CDC ensemble for forecasting COVID-19 hospitalizations.
Opening excerpt (first ~120 words) tap to expand
Article Published: 19 May 2026 An AI system to help scientists write expert-level empirical software Eser Aygün1 na1, Anastasiya Belyaeva2 na1, Gheorghe Comanici1 na1, Marc Coram2 na1, Hao Cui ORCID: orcid.org/0009-0006-2456-083X2 na1, Jake Garrison3 na1, Renee Johnston2 na1, Anton Kast ORCID: orcid.org/0000-0002-0755-99962 na1, Cory Y. McLean ORCID: orcid.org/0000-0001-9928-82162 na1, Peter Norgaard2 na1, Zahra Shamsi2 na1, David Smalling1 na1, James Thompson2 na1, Subhashini Venugopalan ORCID: orcid.org/0000-0003-3729-84562 na1, Brian P. Williams ORCID: orcid.org/0000-0002-2839-01062 na1, Chujun He2,4, Sarah Martinson ORCID: orcid.org/0009-0004-4636-50612,5, Martyna Plomecka2,6, Lai Wei2, Yuchen Zhou2, Qian-Ze Zhu2,5, Matthew Abraham2, Erica Brand2, Anna Bulanova1, Jeffrey A.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Nature.
Discussion
0 commentsMore from Nature
